EPO’s Updated Guidelines Relating to Artificial Intelligence

A preview of the updated guidelines can be found here.

The updated guidelines clarify the patentability of inventions concerning AI and ML. They confirm that these are considered to be computer-implemented mathematical methods. Article 52(2) EPC states that both mathematical methods per se and computer programs (as such), are excluded from patentability.

In order for computer-implemented inventions to be patentable they must solve a technical problem. There is a two step test for them to be considered patentable at the EPO. The first question is: do the claims have technical character? This step is easy to overcome by referring to a computer or a computer-implemented method in the claims.

The second step is more difficult. Here the claims are assessed for inventive step, and if they contain a mix of technical and non-technical features then only features which contribute to the technical character of the invention are taken into account when making that assessment. This means that claims to AI or ML algorithms on their own are not considered to be patentable unless they produce a technical effect. In particular the new guidelines state that the expressions "support vector machine", "reasoning engine" or "neural network" should be “looked at carefully” as they usually refer to abstract models devoid of technical character.

The new guidelines give a number of specific examples of when AI and ML inventions may be patentable, allowing these inventions to be patented in a wide variety of fields. One example given is the use of a neural network in a heart-monitoring apparatus to identify irregular heartbeats. This is considered to make a technical contribution and similar reasoning is likely to apply to all physiological measurements.

The guidelines also refer to use of AI and ML for the classification of documents, stating that an indication of a technical use is required for claims to be seen as technical. The classification of abstract data records likely to be not allowed even if the classification algorithm may be considered to have valuable mathematical properties such as robustness. Classification of digital images, videos, audio or speech signals based on low-level features (e.g. edges or pixel attributes for images) are considered to be patentable. Classification of text documents, however, is likely to be seen as fulfilling a linguistic purpose rather than a technical one.

The guidelines further state that if a classification method serves a technical purpose, the steps of generating the training set and training the classifier may also contribute to the technical character.

Swindell & Pearson is a UK based intellectual property firm that has been helping businesses and individuals obtain patent, trade mark, and design right protection for their ideas, innovations and brands for over 135 years. With its head office in Derby, the firm also has offices throughout the Midlands including in Stoke, Wolverhampton and Stafford, as well as an office in Sheffield. To find out how Swindell & Pearson can help you and your business with any patent, trade mark or design right matters please get in touch via [email protected] or by telephone on 01332 367 051.